DEPS Abstract

Guest >> Sign In

DIRECTED ENERGY PROFESSIONAL SOCIETY

Abstract: 24-Symp-077

UNCLASSIFIED, PUBLIC RELEASE

Phase discontinuity classification from single-aperture irradiance patterns using machine learning.

Shock waves will form over the aperture of aircraft-mounted laser propagation systems due to the acceleration of local flow over the turret body. Accompanying this shock formation is the introduction of a spatial, discontinuous-like increase in the freestream air density, which can be related to the optical phase in propagating light. This shock-induced optical phase discontinuity leads to appreciable beam spreading and a bifurcation of the far-field irradiance pattern, resulting in a sharp reduction in the accuracy and effectiveness of such laser systems. Traditionally, the Shack—Hartmann Wavefront Sensor (SHWFS) is used in tandem with a wavefront reconstruction algorithm to estimate aberrations present in the propagating beam. However, the far-field effects of the shock lead to inaccurate estimation of the induced phase discontinuity, which is especially observed with stronger shocks.
To address this limitation, a shock-tolerant phase reconstruction algorithm was developed. This algorithm identifies the shock-affected information and replaces it with adapted information estimated from heuristic relations between the ratio of the bifurcated irradiance peaks and the induced phase discontinuity. This algorithm was proven to work well over a wide range of flight conditions tested with both simulated and experimental data, however, it lacked versatility due to the aforementioned heuristic relations.
In this study we propose the integration of machine learning into the existing algorithm. A Convolutional Neural Network (CNN) was developed to receive a single aperture observation-plane irradiance pattern and estimate the associated phase discontinuity. Irradiance pattern imagery was generated for a wide range of simulated flight conditions including varying levels of background noise. Testing of the fully trained model shows accurate estimation of the phase-discontinuity across all simulated conditions, with mean absolute errors ranging from 0.0365 to 0.0706 radians. This integration of machine learning into the shock-tolerant phase reconstruction algorithm extends the usefulness of SHWFS into flight conditions with shock waves, enabling robust wavefront sensing in transonic, supersonic, and hypersonic environments.

UNCLASSIFIED, PUBLIC RELEASE